CholecT40 (Cholecystectomy Action Triplet)

CholecT40 是第一个引入的内窥镜数据集，用于研究腹腔镜手术中的细粒度动作识别。它由 40 个腹腔镜胆囊切除术视频组成，并以 <instrument,verb,target> 的形式标注了三重信息。注释跨越 128 个三元组，由 6 类手术器械、8 类动作动词和 19 类手术目标组成。该数据集用作开发深度学习解决方案的基准，以识别三元组形式的手术活动。这是第一次在自然视觉任务中复制与人-物体交互 (HOI) 相同级别的活动识别的外科数据科学工作。父数据集是 CholecT50。

CholecT40 is the first endoscopic dataset developed for research on fine-grained action recognition in laparoscopic surgery. It comprises 40 videos of laparoscopic cholecystectomy procedures, annotated with triplet information formatted as <instrument, verb, target>. The annotations cover 128 triplets, which are composed of 6 categories of surgical instruments, 8 categories of action verbs, and 19 categories of surgical targets. This dataset serves as a benchmark for developing deep learning solutions to recognize surgical activities in the form of triplets. This marks the first work in surgical data science to replicate the same level of human-object interaction (HOI) activity recognition in natural visual tasks. Its parent dataset is CholecT50.

搜集汇总

背景与挑战

背景概述

CholecT40是首个用于腹腔镜手术细粒度动作识别的内窥镜数据集，包含40个胆囊切除术视频和128个<器械,动作,目标>形式的三元组标注，涵盖6类器械、8类动作和19类目标，主要用于开发深度学习解决方案识别手术活动。